Obstacle Detection Method Based on RSU and Vehicle Camera Fusion.

Area of Science:

• Intelligent Transportation Systems
• Computer Vision
• Robotics and Autonomous Systems

Background:

• Road obstacle detection is crucial for intelligent assisted driving technology.
• Existing methods often overlook generalized obstacle detection, impacting safety.
• Generalized obstacles include pseudo-obstacles and objects below the vehicle's maximum passing height.

Purpose of the Study:

• To propose and validate a novel obstacle detection method combining roadside units (RSUs) and vehicle-mounted cameras.
• To enhance generalized obstacle classification accuracy and reduce spatial detection complexity.
• To improve the reliability of obstacle information acquisition in complex driving environments.

Main Methods:

• A hybrid approach integrating a vision-Inertial Measurement Unit (IMU) based method (VIDAR) with an RSU background difference method.
• VIDAR utilizes camera-IMU fusion for object identification and ranging, calculating object height via inverse perspective transformation.
• Data transmission via User Data Protocol (UDP) from vehicle to RSU for pseudo-obstacle removal and error reduction.

Main Results:

• The proposed method achieved accuracy improvements of 2.3%, 17.4%, and 1.8% over compared methods (VIDAR, RSU-based, YOLOv5).
• Obstacle detection speed was enhanced by 1.1% compared to the RSU-based method.
• Demonstrated expansion of detection range and effective elimination of false obstacle information.

Conclusions:

• The fused RSU-vehicle camera approach significantly improves generalized road obstacle detection.
• VIDAR technology effectively addresses challenges in acquiring accurate obstacle information.
• The method offers a robust and efficient solution for enhancing road safety in intelligent assisted driving.